I built a web tool to see and edit what an AI thinks before it answers
I run a small AI lab and playground and got super excited about Anthropics paper "Verbalizable Representations Form a Global Workspace in Language Models" (https://transformer-circuits.pub/2026/workspace/index.html) It talks about how they use a tool they call a Jacobian Lens to view inside the middle layers of LLM while it's working before it commits to a word (token). I wanted to see if I could get a version of this running on the open models and to my surprise it worked! I ran some experiments with it and build a public facing free tool anyone can use with your…
In plain words
This tool lets users view and modify what language models are thinking in their internal layers before generating answers. Built on research from Anthropic's Jacobian Lens technique, it applies this interpretability method to open-source models. Users can watch how different word possibilities activate during processing—for example, seeing "ocean," "sea," and "surf" light up before the model settles on "waves." The tool also allows direct editing of internal states to see how changes propagate through the model's response. It's free and designed for anyone interested in understanding how AI systems form their outputs.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I run a small AI lab and playground and got super excited about Anthropics paper "Verbalizable Representations Form a Global Workspace in Language Models" (https://transformer-circuits.pub/2026/workspace/index.html) It talks about how they use a tool they call a Jacobian Lens to view inside the middle layers of LLM while it's working before it commits to a word (token). I wanted to see if I could get a version of this running on the open models and to my surprise it worked! I ran some experiments with it and build a public facing free tool anyone can use with your own prompts. Ask the model to describe a symbol of "three curving lines of water" and you can watch "ocean", "sea", and "surf" light up a few layers deeper before it settles on "waves". You can also edit the internal state. Insert "fire" into the middle layer of the ocean prompt and the answer shifts to something about heat. For fun / curiosity sake, I also developed way to let the model read its own inner workspace and then decide to suppress or amplify a concept, and run the prompt again. Interesting finding from running it across models. J-lens beats a plain logit lens on some architectures and does nothing on others, and it isn't about size. A 0.5B Qwen reads better than a 2.8B Pythia. Every Pythia I tried gained basically nothing; the Llama and Qwen models gained a lot. https://lucid.earthpilot.ai/research This is a 48 hour old project based on emerging research and built on a small model, a small probe set on rented GPUs - but I found it genuinely exciting. The code is open. I also included a page context "Docent" AI agent you can chat with about whatever you see to help understand what is going on. Happy to have folks poke around and break it. I imagine the applications for allowing models to self-reflect / edit internal states can be useful for alignment, confidence, bias detection, etc. and this tool lets you play with the early stages of that.
Does the same job
all alternatives →- L3Llama 3.2 Interpretability with Sparse Autoencoders2024 · github.com · ▲579
I spent a lot of time and money on this rather big side project of mine that attempts to replicate the mechanistic interpretability research on proprietary LLMs that was quite popular this year and produced great research papers by Anthropic [1], OpenAI [2] and Deepmind [3]. I am quite proud of this project and since I consider myself the target audience for HackerNews did I think that maybe some of you would appreciate this open research replication as well. Happy to answer any questions or face any feedback. Cheers [1]…
- IBI built a tiny LLM to demystify how language models workApr 2026 · github.com · ▲915
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
- CMCreate mind maps to learn new things using AI2024 · github.com · ▲170
Enter a topic and get a learning mind map generated by an LLM with links to learn more about each subtopic. You can use it with local models (through Ollama) or external models. If you have any feedback, please share it! Hope it's useful Demo: https://youtu.be/Y-9He-tG3aM
ThoughtDAG – An editable context graph for LLM conversations22d ago · chenxiachan.github.io · ▲136ThoughtDAG indexes local agent conversations across tools, finds the turns relevant to your work, and turns them into editable context graphs.
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, July 2026
the whole month →- IR
I might be the only SRE on Earth with his own bowling center. It's a more in-depth gig than you'd think. My family and I bought an abandoned 8-lane bowling center in the rural mid-west. In our small town there weren't many recreation options for families. You've heard of a food desert? This is an R&R desert. It had been abandoned for a good reason. The roof leaks, the electrical system was constantly surging, and my 70-year-old bowling equipment (still) doesn't work perfectly. The system that keeps your score is particularly interesting to me. It's the thing you watch during your game, but…
Life & fun · Jul 2026
- EElevators▲1,680
Life & fun · Jul 2026 · john.fun
- 1W18 Words▲1,160
Life & fun · Jul 2026 · 18words.com
- BA
Over the past few months, our team has been building more and more slidedecks using web frontend technologies with coding harnesses like Claude Code, but a common complaint is to make even small edits we need to edit the code either manually or via the harness. To avoid this loop, I ended up creating Bento, a single HTML file with everything you need in a slide tool including animations and shared editing. There's no install or cloud login, everything works offline. The default deck is around 560 KB and it doesn't need to fetch anything once you got it. Open it in a browser and then you can…
Dev tools · Jul 2026 · bento.page
- GG
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me. But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility. I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context.…
AI · Jul 2026 · github.com
